Open source · MIT · Built by Arthur
Deckard scans your endpoints, cloud and network for the AI agents nobody registered, then hands every one of them to your SIEM. Read-only. Runs in your environment.
Download https://raw.githubusercontent.com/arthur-ai/deckard/main/docs/agent-setup/prompt.md with curl, read all of it, and follow its instructions to set up Arthur Deckard on this machine.
Download https://raw.githubusercontent.com/arthur-ai/deckard/main/docs/agent-setup/prompt.md with curl, read all of it, and follow its instructions to set up Arthur Deckard on this machine.
git clone https://github.com/arthur-ai/deckard.gitcd deckard/docker-compose./scan-local.shdocker compose up -d --wait
01 / Get running
Into Claude Code or Codex, from any folder. Prefer doing it by hand? The terminal tab has the four commands.
Clones the repo, checks Docker and osquery, and runs the local scan. If anything needs installing or your password, it stops and tells you the command. It never runs sudo.
Deckard sends each agent to an Elasticsearch running beside it, and Kibana’s Discover lists them. Then it can walk you through adding your cloud, MDM or SIEM.
curl -s -u elastic:changeme 'localhost:9200/_query?format=txt' …@timestamp | agent.name | source.name------------------------+----------------+---------------2026-10-26T21:06:33.120Z|Claude Code |this-machine2026-10-26T21:06:33.120Z|Claude Desktop |this-machine2026-10-26T21:06:33.120Z|Cursor |this-machine...
02 / Scan your entire organization
01 / 05
Deckard includes a comprehensive set of osquery signatures, the leading endpoint scanner used by security teams everywhere. These signatures are constantly updated by our agent software factory. They can be easily deployed by organizations using popular fleet and mobile device management platforms.
02 / 05
A robust set of integrations is available, with leading cloud providers including Google Vertex, AWS Bedrock, and Azure Agent Foundry.
03 / 05
Deckard connects to your SIEM and searches your existing logs for signs of agent activity, using data you already collect.
04 / 05
Deckard uses network logs to identify agents using LLMs and other agentic resources.
05 / 05
Deckard contains support for Managed Agent Environments such as OpenAI Agents API and Claude Managed Agents. This ensures as new agents are brought online, they are quickly identified and tracked.
Sends to Splunk HECany HTTPS webhook
Missing one you use? Open an issue03 / Events
agent.external_id and alert on what matters.
{
"event_type": "arthur.discovery.agent",
"schema_version": 1,
"source": { "name": "bedrock-prod", "vendor": "aws_bedrock" },
"agent": {
"external_id": "arn:aws:bedrock:us-east-1:…:agent/AB12",
"name": "inventory-agent",
"last_seen": "2026-10-07T17:58:12+00:00"
}
}new agent.external_idAn agent nobody has seen before.
outcome.succeeded: falseA source Deckard can no longer read, and why.
outcome.denied_scopesA project or account the scan was refused, so its agents are invisible.
device_coverage.unreadable_devicesMachines the scan couldn’t read. That’s where unmanaged AI sits.
Deckard only lists and searches. Each source needs read permission, and credentials stay out of every log and event.
One container, no database. It talks only to your sources and your destination. Your inventory never leaves your network.
Every scan sends what the source reports now. A restarted Deckard just scans again.
A source that fails or an account it was refused raises its own alert instead of going quiet.
Deckard is the standalone discovery mode of the Arthur Engine. The Arthur Platform runs the same scans, then adds triage for every agent found, plus policies, monitoring and guardrails for the ones you bring under governance.